This dissertation investigates three types of mesoscale (milli- and microscale) magnetic robots: magnetic swimmers, magnetic modular cubes, and ferromagnetic micro-particles. Mesoscale magnetic robots show great potential for revolutionizing many aspects of medical and clinical applications. The dissertation investigates Millimeter-scale Magnetic Rotating Swimmers (MMRSs) that could be used to improve surgical procedures. An external rotating magnetic field produces a torque on the swimmers to make them rotate. MMRSs have propeller fins that convert the rotating motion into forward propulsion. The dissertation reports on optimization studies for the MMRS designs and control techniques used experimentally to remove thrombi from a bifurcating vascular model. The dissertation also presents data-driven models to improve MMRS’s 3D path-following performance of a time-delayed sensing system. An algorithm for 2.5D closed-loop control of the MMRS using only 2D ultrasound feedback is proposed and tested experimentally. In addition, a preliminary study of the biocompatibility of the MMRSs is presented. Magnetic Modular Cubes (MMCs) are scalable modular subunits with embedded permanent magnets in a 3D-printed cubic body. Due to the MMC’s cubic design, magnetically connected structures of MMCs are polyominoes in 2D and polycubes in 3D. MMCs represent progress toward a mesoscale manufacturing method controllable by an external force field that combines the precision of modules, the reusability of Legos, and the self-assembly of DNA. The dissertation provides a family of designs of MMCs and a 2D low-fidelity motion planner that computes all reachable polyomino shapes (and their shortest movement sequences) from an arbitrary initial configuration. A closed-loop control method is presented for self-assembling the MMCs in 2D using computer vision-based feedback with re-planning techniques. Furthermore, methods to enumerate polyominoes and polycubes are presented. For biomedical applications in targeted therapy delivery and interventions, a large swarm of micro-scale particles has to be moved through a maze-like environment to a target region. The dissertation demonstrates how to use a time-varying magnetic field to gather ferromagnetic micro-particles to a desired location using reinforcement learning. In addition, methods to overcome the simulation-to-reality gap are explained in the dissertation.